Cognitive OS Starter Kit
Download four Markdown files for building agent memory, role cards, decision logs, and weekly review loops.
Cognitive OS Starter Kit
A minimal set of Markdown files for building an AI-native operating system around memory, agents, workflows, and human judgment.
Most people try to improve AI work by collecting prompts or testing more tools. The Cognitive OS approach starts somewhere else: make the context around intelligence durable, inspectable, and reusable.
The Starter Kit gives you four plain-text files you can copy into any workspace.
What you get
1. Starter filesystem
A small folder structure for memory, workflows, agents, projects, and reviews.
Download the starter filesystem
2. Agent role card
A reusable file for defining what an agent is responsible for, what it must read, what it may change, what it must not do without approval, and what standard its output must meet.
3. Decision log
A compact record for decisions that future humans and agents should not rediscover from scratch.
4. Weekly agent review
A weekly loop for turning agent activity into compounding memory, constraints, workflow improvements, and better role cards.
Download the weekly review template
Download the whole kit
Start with the README, then copy the four files into your own workspace.
Download the Starter Kit README
How to use it in 30 minutes
- Create a
/cognitive-osfolder in your normal working directory. - Copy the starter filesystem into it.
- Pick one recurring agent role: researcher, editor, operator, reviewer, analyst, or meeting summarizer.
- Fill out one role card for that agent.
- Add one decision to the decision log.
- Schedule one weekly review.
Do not try to design the perfect system on day one. The point is to create enough structure that agents can read the same context humans use.
Why this matters
Agents become more useful when they are surrounded by durable context:
- what the work is for
- what has already been decided
- what files define the workflow
- what good output looks like
- what requires human judgment
- what should be updated after the work is done
This is the difference between using AI as a chat window and operating with a Cognitive OS.